InG (Intention-Grounded Gestures)

该数据集名为InG，它通过将结构化的、关注意图的注释与语言功能关联起来，补充了BEAT-2数据集，这些注释将语言功能与手势行为相联系，使得能够生成可控意图的手势。该数据集包含了16种交流功能类型的注释，覆盖了如强调、指示、心理状态和过程等语用类别。其规模包括34,641个训练样本、3,598个验证样本和9,674个测试样本。该数据集的任务是基于交流意图的手势生成。

The dataset named InG supplements the BEAT-2 dataset by associating structured, intention-focused annotations with linguistic functions. These annotations link linguistic functions to gestural behaviors, enabling the generation of gestures with controllable communicative intentions. This dataset includes annotations for 16 types of communicative functions, covering pragmatic categories such as emphasis, deixis, mental states and processes. In terms of scale, it contains 34,641 training samples, 3,598 validation samples and 9,674 test samples. The task of this dataset is gesture generation conditioned on communicative intentions.